Power line alarm method and system with AI function

Abstract

The invention relates to the technical field of electric power, in particular to an electric power circuit alarm method and system with an AI function, and the method comprises the steps: collecting multi-source data, and carrying out the time-space calibration of the information of the multi-source data through a Kalman filtering algorithm; constructing a multi-physics field coupling prediction model, and calculating standardized scores of different risk types; different types of risk feature weights are calculated by using a quantum-graph attention network fusion model, and fault severity scores corresponding to different types of risks are calculated through a migration reinforcement learning fault diagnosis model; constructing a comprehensive risk value calculation formula by combining the standardized scores of the different risk types, the feature weights of the different types of risks and the fault severity scores corresponding to the different types of risks, and calculating a comprehensive risk value; and carrying out grade division on the risk according to a threshold interval in which the comprehensive risk value is located, and pushing different early warning information and processing strategies to different responsible personnel according to different risk grades. According to the scheme, accurate early warning is realized through multi-source data space-time calibration and AI fusion.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a power line alarm method and system with AI function. Background Art

[0002] With the intelligent upgrade of power grids and increased electricity demand from consumers, power systems are placing higher demands on risk monitoring and operational efficiency. However, traditional power line operation and maintenance (O&M) face significant technical bottlenecks: First, multi-source monitoring data is severely fragmented. Data collected by sensors such as laser ranging and fiber-optic temperature measurement lack a unified spatiotemporal benchmark, resulting in millisecond-level time errors and spatial positioning errors exceeding ±5 meters. This leads to significant deviations in risk attribution. For example, lightning and wildfire data often misjudge correlations due to spatiotemporal misalignment. Second, physical mechanism analysis is insufficient. Traditional single models only consider a single risk factor and fail to capture the effects of multi-physics coupling. This results in prediction errors exceeding 20% ​​in complex scenarios, making it prone to underreporting significant risks. Third, intelligent decision-making capabilities lag. Traditional AI algorithms lack cross-domain adaptability, resulting in prediction errors exceeding 40% in new scenarios or rare operating conditions. Furthermore, they rely on fixed thresholds and cannot dynamically optimize strategies, resulting in manual intervention costs exceeding 30% and inefficient emergency response. These issues severely hinder the safety and reliability of power grid operations. Innovative technical solutions are urgently needed to overcome existing technical bottlenecks, improve the accuracy of risk warnings, and enhance the safety of power grid operations. Summary of the Invention

[0003] This invention achieves intelligent and accurate prediction of power line risks by integrating multi-source data spatiotemporal calibration, multi-physics field modeling and AI algorithms (Q-GAT, DRL).

[0004] The technical solution proposed by the present invention is: a power line alarm method with AI function, the method comprising: Use multiple sensors to collect multi-source data, receive multi-source data information through the Beidou dual-mode module, and use the Kalman filter algorithm to perform spatiotemporal calibration on the multi-source data information; Based on the collected multi-source data information, a multi-physics field coupling prediction model is constructed, and standardized scores for different risk types are calculated through the multi-physics field coupling prediction model. The multi-physics field coupling prediction model includes a conductor thermal-mechanical coupling model, a wildfire spread dynamic model, a lightning risk prediction model, a conductor dancing prediction model, a bird activity risk model, a dust and pollution flashover risk model, a geological disaster risk model, and an equipment aging risk model. The quantum-graph attention network fusion model is used to calculate the weights of different risk features, and the fault severity scores corresponding to different types of risks are calculated through the transfer reinforcement learning fault diagnosis model; A comprehensive risk value calculation formula is constructed by combining the standardized scores of different risk types, the weights of different risk characteristics, and the fault severity scores corresponding to different risk types. The comprehensive risk value is calculated using the comprehensive risk value calculation formula. Risks are classified into different levels according to the threshold range of the comprehensive risk value, and different warning information and processing strategies are pushed to different responsible personnel according to different risk levels.

[0005] Preferably, the spatiotemporal calibration comprises the following steps: The B1 and B3 frequency signals are received through the Beidou dual-mode module. The Kalman filter algorithm is used to deduct the error between the sensor's original time and the Beidou standard time, the ionospheric delay correction value, and the tropospheric delay correction value to achieve time calibration. The positioning error vector of the sensor's original spatial coordinates is corrected to achieve spatial calibration. The improved DTW algorithm is used to introduce time and space constraint factors to achieve data alignment.

[0006] Preferably, the conductor thermo-mechanical coupling model constructs a model formula by collaboratively predicting changes in temperature, stress and ice thickness, divides the unit conductor into 100 units, and uses the finite element method to solve it; the wildfire spread dynamic model uses the principle of heat conduction and fire risk assessment indicators to analyze the influence of surface temperature, vegetation humidity, and wind speed on the fire spread speed, and uses the Kalman filter algorithm to update the model parameters; the lightning risk prediction model establishes an equipment damage probability model through a Poisson process combined with the equipment tolerance current threshold to predict the probability of equipment damage caused by lightning strikes; the conductor galloping prediction model is based on fluid mechanics and vibration theory, analyzes the coupling relationship between wind speed, wind direction and the natural frequency of the conductor, and establishes an instability coefficient model for conductor galloping.

[0007] Preferably, the bird activity risk model is based on image recognition and statistical analysis, sets the safe distance threshold between bird nests and conductors and the bird activity density threshold, and establishes a bird damage risk assessment model; the dust and pollution flashover risk model is based on the flashover mechanism of polluted insulators, combined with The geological disaster risk model is based on the principle of coupling between surface displacement and acceleration, and a dual-parameter risk assessment model of surface displacement rate and acceleration peak is established; the equipment aging risk model is based on the theory of material fatigue and corrosion, and a dual-index assessment model of insulator crack extension and hardware corrosion is established.

[0008] Preferably, the process of obtaining weights of different types of risk characteristics is as follows: The physical field characteristics of multi-source data are mapped into quantum states, and superposition states and entangled states are generated through quantum gate operations to form a quantum feature space; the quantum states are measured, nonlinear features are extracted, and enhanced feature vectors are generated; the power network is abstracted into a graph structure, where nodes represent poles or equipment, and edges represent physical connections or risk propagation relationships between nodes. The association weights between nodes are calculated through the attention mechanism; the neighbor node features are aggregated based on the attention weights, the current node representation is updated, and the weights of different types of risk features are output.

[0009] Preferably, the transfer reinforcement learning fault diagnosis model processing process is as follows: The operating conditions of power lines are converted into a computable state space, and a set of fault handling actions is preset for fault handling. A deep neural network is used to construct a policy function, which inputs the state, outputs the action probability distribution, sets reward rules, updates the policy network parameters based on the policy gradient algorithm, and outputs the fault type decision probability matrix and the fault severity score vector. When the system accesses new regional data, the policy network parameters of the source domain and the target domain are fused through domain adaptation technology.

[0010] Preferably, the specific formula for calculating the comprehensive risk value is as follows: ; in: is the comprehensive risk value; For the The combined weight of the risk class; For the Standardized scoring of class risks; is the risk coupling coefficient; For risk and The correlation coefficient of The output of the Q-GAT model is Class risk characteristic weights; is the regulating factor; The output of the transfer reinforcement learning model Fault severity score corresponding to the risk class; For risk category.

[0011] Preferably, the specific process of the level division is as follows: when When the risk level is low , archive monitoring data, carry out inspections according to normal requirements, and regularly evaluate and maintain the system's performance; When the risk level is medium , the system will automatically generate routine maintenance work orders, arrange operation and maintenance personnel to conduct inspections and maintenance as planned, strengthen online monitoring, and increase the frequency of data collection; When the risk level is high The system pushes AR maintenance guides to operation and maintenance personnel, guides on-site maintenance through augmented reality technology, activates drone inspections, increases inspection frequency, closely monitors line operation status, adjusts operating parameters in real time, and conducts local power outage inspections when necessary; When the risk level is urgent The system prioritizes the use of 5G slicing networks to transmit monitoring data, dynamically adjusts line protection strategies, reduces transmission capacity, and arranges professional operation and maintenance personnel to rush to the site with portable testing equipment to conduct emergency investigations and preliminary processing; When the risk level is catastrophic The system automatically triggers the tripping protection, cuts off the power supply of the faulty line, starts satellite relay communication, dispatches drone fire-fighting clusters and repair robots, and notifies the emergency command center to organize a cross-regional professional repair team to carry out large-scale rescue and disaster relief work.

[0012] The present invention also provides a power line alarm system with AI function, which is used to execute the power line alarm method with AI function.

[0013] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the power line alarm method with AI function.

[0014] Beneficial effects of the present invention: This solution builds a high-density monitoring network using multiple sensors, including laser ranging, satellite remote sensing, and fiber-optic temperature measurement. This network covers eight risk categories, including ice cover, wildfires, and lightning, addressing the problem of missed detections by a single sensor. Leveraging the BeiDou dual-mode module (B1 / B3 frequencies) and the Kalman filter algorithm, it achieves microsecond-level time calibration (±500ns accuracy) and meter-level spatial calibration (positioning error vector correction). An improved DTW algorithm introduces spatiotemporal constraints to ensure microsecond-level alignment of multi-source data. For example, after calibration, data from laser ranging sensors (±1mm accuracy) and distributed fiber-optic temperature measurement systems (temperature gradient ±0.5°C / 100m) can accurately correlate ice thickness and temperature changes on the same tower at the same time, avoiding data mismatches caused by spatiotemporal bias. This technology reduces the error rate of risk tracing to less than 3%, providing a reliable data foundation for subsequent physical models and intelligent algorithms, and establishing a "spatiotemporally accurate digital twin" of power lines, ensuring that monitoring results truly reflect on-site conditions.

[0015] This solution integrates physical principles with quantum computing technology to construct a multi-physics coupling model (such as a conductor thermo-mechanical coupling model). Using the finite element method, the conductor is divided into 100 units, analyzing the physical mechanisms of risks such as icing and lightning strikes from the perspective of energy conservation. Furthermore, a quantum-graph attention network (Q-GAT) is introduced to map multi-source data into quantum states. Quantum circuits are then used to concurrently extract high-dimensional features (such as quantum correlations between icing rate, wind speed, and tower stress). The graph attention mechanism is then used to analyze the risk propagation relationships between nodes (towers) in the power grid topology. For example, in wildfire risk prediction, the physical model uses the combustion heat release rate and fire danger index to distinguish between "high temperature without vegetation" and "real fire" scenarios. Q-GAT, by learning from satellite thermal imagery data, exploits the nonlinear coupling characteristics of surface temperature and vegetation moisture, achieving a fire location error of less than 100 meters and improving prediction accuracy to 95%. The two work together to achieve a closed loop where "physical laws constrain data-driven development, while algorithmic optimization feeds back to model accuracy," breaking through the bottleneck of traditional models in modeling complex coupled risks.

[0016] This solution introduces a transfer reinforcement learning algorithm, which uses fault characteristics (such as ice accumulation rate and lightning current peak) as states, optimizes the decision-making strategy through the policy gradient algorithm, and outputs the fault type probability matrix and severity score. When the system enters a new area or faces new working conditions, the policy network parameters of the source domain (such as plain areas) and the target domain (such as plateau areas) are integrated through domain adaptation technology to quickly adapt to environmental differences. For example, in the icing scenario of the Sichuan-Tibet line, DRL can improve the learning efficiency of the ice melting strategy by 60% by migrating the de-icing experience of North China with only 50 sets of local data, and dynamically adjust the response level (such as This technology upgrades from "threshold-triggered passive response" to "algorithm-driven proactive decision-making," shortening the response time for new risks (such as rare bird nesting behavior) from 24 hours to 2 hours. It also optimizes resource allocation (for example, prioritizing drone inspections for high-risk R3 levels), reducing operation and maintenance costs by over 30%. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a power line alarm method with AI function according to the present invention; Figure 2 This is a hierarchical early warning flow chart of a power line alarm method with AI function according to the present invention. DETAILED DESCRIPTION

[0018] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are for illustrative purposes only, and those skilled in the art will readily appreciate other obvious variations. The basic principles of the present invention defined in the following description may be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the present invention.

[0019] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the elements may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0020] like Figure 1 and Figure 2 As shown, multi-source data collection and preprocessing are first performed. Distributed fiber optic temperature sensors are deployed every 500 meters and on each tower. Laser ranging sensors measure the conductor ice thickness in real time (with an accuracy of ±1mm). Distributed fiber temperature measurement (DTS) is used to obtain the conductor temperature gradient (with an accuracy of ±0.5°C / 100m). This real-time data triggers the de-icing device and load adjustment. Unmanned aerial vehicles (UAVs) are used to inspect surface temperatures using infrared thermal imagers (640×480 resolution). Fire points are monitored using satellite remote sensing data (10-meter resolution). Warnings are triggered when the temperature is ≥80°C or the smoke concentration is >500 ppm. Detection substations and a central station are deployed to form a lightning location system. This system (with an accuracy of ±500 meters) tracks the location and intensity of lightning strikes in real time, counts the number of lightning strikes per hour, and records the peak current (kA). This system, combined with conductor temperature and sag data, predicts lightning strike risks. Wind speed and direction are monitored using ultrasonic anemometers (accuracy ±0.1 m / s), and conductor vibration frequency (2-20 Hz) is collected using accelerometers (accuracy ±5% FS). When wind speeds ≥ 15 m / s or the vibration frequency exceeds a threshold (e.g., 15 Hz), warnings are issued regarding the risk of galloping, allowing for dynamic adjustments to transmission capacity. Drones equipped with LiDAR (accuracy ±0.1 m) monitor tree heights along designated inspection routes quarterly. AI-powered video analysis analyzes the vertical distance to the tree line, triggering warnings when the distance is less than 5 meters or the growth rate exceeds a threshold (e.g., 10 cm / month). High-definition cameras (4K resolution) coupled with the YOLOv5 algorithm analyze video streams in real time to identify bird flocks (>10 birds) and nesting behavior (<2 meters from the conductor), preventing short circuits caused by bird damage. These surveillance cameras are primarily deployed in wetlands and along bird migration routes. Laser dust sensors and hygrometers are deployed every kilometer in industrial and coastal areas. Laser dust sensors (detecting 0.3 Particulate matter) monitors PM2.5 concentration (>300 ), hygrometers (accuracy ±2%RH) simultaneously collect air humidity and calculate insulator salt density (>0.1mg / cm² to assess flashover risk). Tipping bucket rain gauges (accuracy ±2%) are deployed every 500 meters in river valleys and low-lying areas, recording rainfall in real time (providing warnings when rainfall exceeds 50mm / h). Inclination sensors (accuracy ±0.1°) are installed on towers to monitor the inclination angle (providing warnings for foundation scour risks when the inclination exceeds 3°). Fiber Bragg grating temperature sensors and temperature and humidity transmitters are installed on all towers. Fiber Bragg grating temperature sensors (accuracy ±0.3°C) collect conductor temperature (providing warnings when the temperature exceeds 85°C or is less than -40°C). Temperature and humidity transmitters (accuracy ±2°C / ±3%RH) monitor environmental parameters, dynamically adjusting transmission capacity or initiating anti-freeze measures based on the monitoring results. Smart cameras are deployed around construction sites in cities and suburbs. Using AI visual recognition technology (e.g., cranes and excavators), these cameras analyze equipment type and distance (triggering audible and visual alarms when the distance is less than 50 meters) and record operation duration (increasing to an alert if the distance is greater than 30 minutes). Monitoring equipment is deployed on every tower in remote areas, including microwave radar. The radar (with a detection range of 100 meters) detects abnormal human movement (speeds greater than 5m / s) and uses AI-powered video to analyze climbing behavior. Rogowski coils and ultra-high frequency (UHF) sensors are deployed at the outgoing line of substations using an encrypted method. Rogowski coils (with an accuracy of 0.1%) collect real-time conductor current (initiating an overload alert when the current exceeds 1.2 times the rated value). Ultra-high frequency (UHF) sensors (300-3000MHz) detect partial discharge (initiating an insulation defect alert when the current exceeds 10pC). Monitoring points are deployed every 500 meters around the chemical park, each equipped with a gas sensor. These sensors (SO2 / NO2 detection accuracy ±5% FS) continuously monitor acid rain composition and corrosive gas concentrations (alerts are issued when SO2 > 50 ppm or pH < 4.5). Accelerometers (accuracy ±0.1 mg) and GNSS modules (positioning accuracy ±1 cm) are deployed every 50-100 meters in landslide-prone areas to monitor ground displacement in real time. Ultrasonic flaw detectors (resolution 0.1 mm) are installed on the insulator strings of each tower to detect internal cracks, and corrosion sensors (accuracy ±5% metal thickness loss) are placed on the surfaces of hardware. The density of these sensors can be adjusted based on actual conditions.

[0021] Risks in power lines are diverse and originate from different sources, making comprehensive coverage impossible with a single sensor. For example, measuring temperature alone cannot predict wildfires, and measuring wind speed alone cannot determine equipment degradation. Multi-source data is like a "multi-dimensional physical examination report." By collecting data from different dimensions, we ensure that no potential risks are missed and achieve comprehensive protection for power lines. By analyzing multi-source data, we can predict potential risks in advance, preventing the spread of disasters and creating larger ones, thus nipping disaster damage in the bud.

[0022] Various sensors receive B1 and B3 frequency signals through the BeiDou dual-mode module and use the Kalman filter algorithm to calculate time and space data: The time calibration formula is as follows: ; in: is the time after calibration; Beidou Standard Time; is the original time of the sensor; is the ionospheric delay correction value, calculated by dual-frequency GPS, with a range of 0-500ns; is the tropospheric delay correction value, calculated using the Saastamoinen model, with a typical value of 2-20 .

[0023] The spatial calibration formula is as follows: ; in: is the spatial coordinate after calibration; is the original space coordinate of the sensor; is the positioning error vector.

[0024] Data alignment: Improved DTW algorithm introduces spatiotemporal constraints (value less than 1) to achieve microsecond-level time alignment. The formula is as follows: ; in: is the cumulative distance; is the Euclidean distance of the feature vector.

[0025] Because data collected by multiple sensors is inconsistent in time and space, data calculation errors can occur. For example, mistakenly associating ice cover data from Tower A with Tower B could trigger false alarms. A unified spatiotemporal reference is fundamental to data analysis, much like aligning the edges of a puzzle. BeiDou standard time is used to correct for errors in the sensor's raw time, deducting ionospheric and tropospheric delays to ensure that all data is in the same time base (with microsecond accuracy). This process can be described in the "Data Alignment" section. Sensor positioning errors are corrected using BeiDou coordinates to ensure accurate tower position data. This process can be described in the "Map Correction" section. Different sensors have different sampling frequencies. Using a dynamic time warping algorithm, spatiotemporal constraints are introduced to align data in chronological order. This process can be described in the "Movie Frame Synchronization" section. Spatiotemporal calibration and data alignment ensure accurate correspondence between multiple data sources at the same time and location, improving model calculation reliability. Calibrated data can also accurately trace the time and space points at which risks occurred, facilitating subsequent analysis.

[0026] Based on the collected multi-source data, a multi-physics coupling prediction model is constructed. This model includes a conductor thermal-mechanical coupling model, a wildfire spread dynamic model, a lightning risk prediction model, a conductor dancing prediction model, a bird activity risk model, and a dust and pollution flashover risk model. The innovative intelligent algorithm cluster and the multi-physics coupling prediction model complement and work together. The multi-physics coupling prediction model is based on physical principles and, drawing on the actual physical processes of power line operation, constructs a mathematical model to describe and predict various risks. The innovative intelligent algorithm cluster, on the other hand, uses a data-driven approach to deeply mine and analyze data using methods such as the Quantum-Graph Attention Network (Q-GAT) fusion model and a transfer reinforcement learning fault diagnosis model. In practical applications, the multi-physics coupling prediction model provides physical explanations and constraints for the innovative intelligent algorithm, ensuring that its predictions are more consistent with actual physical laws. The innovative intelligent algorithm cluster can also handle complex nonlinear relationships and high-dimensional data that are difficult for the multi-physics coupling model to resolve. By learning from large amounts of historical data, it can discover hidden patterns and regularities, further optimizing risk prediction results. The combination of the two improves the comprehensiveness and accuracy of risk predictions. The Q-GAT fusion model leverages the parallel processing capabilities of quantum computing and the structural feature extraction advantages of graph attention networks to deeply extract key features and underlying relationships from high-dimensional, complex spatiotemporal data. When processing power network topology data, it not only captures the characteristics of individual nodes (such as towers) but also mines the correlations between them, enabling a more comprehensive understanding of the operational status of power lines and providing richer, more accurate feature information for risk prediction. The transfer reinforcement learning fault diagnosis model continuously interacts with the environment, using fault characteristics as its state and autonomously learning and optimizing fault response strategies based on reward feedback. When faced with new fault scenarios or data changes, it can quickly adapt and automatically adjust decisions, improving the accuracy and efficiency of fault diagnosis. This shift from passive response to active decision-making makes the system more intelligent and adaptive.

[0027] The conductor thermal-mechanical coupling model is shown below: ; ; ; in: is the wire current, and the current passing through the wire generates Joule heat; is the wire resistance, which is related to the wire material, length, and cross-sectional area; The latent heat of ice phase change and its thermal impact on the conductor; Heat exchange for wind power; The instantaneous heat generated by the lightning strike; is the convection heat transfer coefficient, which reflects the heat exchange capacity between the conductor and the air; is the surface area of ​​the conductor; is the conductor temperature; is the ambient temperature; is the surface emissivity of the conductor; is the Stefan-Boltzmann constant; is the wire extension; The gravity of ice cover; For wind power; is the impact force of lightning strike; is the original length of the wire; is the elastic modulus of the conductor material; is the linear expansion coefficient of the conductor material; is the ice cover growth rate; For safe ice cover growth rate; is the critical ice growth rate; is the conductor stress; is safety stress; is the critical stress; is the conductor temperature; is the optimal temperature; is the maximum temperature; is the lowest temperature; Provides a standardized score for conductor thermal risk.

[0028] This formula divides a unit conductor into 100 units and is solved using the finite element method. In an icing environment, this formula can collaboratively predict changes in temperature, stress, and ice thickness with an error rate of less than 3%. Combining energy balance relationships such as current heat generation, ice phase change latent heat, and wind heat dissipation, as well as the mechanical deformation principles of conductors under external forces such as icing, wind, and lightning strikes, a comprehensive model is established to analyze conductor temperature and stress changes. The conductor is divided into multiple units, and the finite element method is used to simulate the state of different units under the influence of parameters such as current, temperature, and ice thickness. Real-time monitored current values, ambient temperature, and ice growth rate are input, and calculations are performed based on the elastic and thermal expansion characteristics of the conductor material. The dynamic changes in conductor temperature, stress, and ice thickness are accurately predicted with an error rate of less than 3%, providing a scientific basis for ice melting operations and transmission load adjustments.

[0029] The wildfire spread dynamic model is as follows: ; ; in: is the rate of change of temperature with time; is the Hamiltonian operator, describing the spatial changes; is the thermal conductivity coefficient; is the combustion heat release rate; is the fire danger index ( ,in is the surface temperature, is vegetation humidity, is wind speed); is the lowest fire danger index; is the highest fire danger index; A standardized score for wildfire spread risk.

[0030] Integrating drone thermal imaging data, the ensemble Kalman filter (EKF) updates model parameters every 10 minutes, achieving dynamic prediction of fire location with an error of less than 100 meters. Based on the principles of heat conduction and fire risk assessment indicators, the system analyzes the impact of factors such as surface temperature, vegetation moisture, and wind speed on fire spread, and establishes a fire spread prediction model. Using drone thermal imaging to obtain surface temperature data, combined with satellite remote sensing information on vegetation moisture and wind speed, the Kalman filter algorithm updates model parameters in real time, dynamically adjusting the predicted fire location. This system achieves dynamic tracking of fire location with an error of less than 100 meters. When the surface temperature exceeds 80°C or smoke concentration exceeds the standard, a wildfire warning is immediately triggered.

[0031] The lightning risk prediction model is as follows: The Poisson process formula is used to describe the probability of lightning strikes, and the risk of lightning to line equipment is evaluated in combination with the equipment damage probability model. The Poisson process formula is as follows: ; The equipment damage probability model formula is as follows: ; ; in: For time Internal The probability of a lightning strike; is the average number of lightning strikes per unit time; is the number of lightning strikes; is the probability of equipment damage; is the adjustment coefficient of the equipment damage probability curve; is the actual lightning current; The device withstand current threshold; is the lightning strike density; is the lowest lightning strike density; is the highest lightning strike density; is the coefficient of variation of lightning current intensity; is the minimum coefficient of variation; is the maximum coefficient of variation; A standardized score for lightning strike risk.

[0032] Using the Poisson process to describe the probability of lightning strikes, combined with the equipment's current tolerance threshold, a lightning damage risk model for line equipment was established. The lightning location system counted the number of lightning strikes and current peaks within the area, analyzed lightning strike density and current fluctuations, and assessed the impact risk of lightning strikes on conductors and equipment. This quantified lightning risk level and predicted the probability of equipment damage caused by lightning strikes, providing data support for optimizing lightning protection measures.

[0033] The wire galloping prediction model formula is as follows: ; ; in: The instability coefficient is used to evaluate the possibility of conductor galloping. The larger the value, the higher the galloping risk. is the air density; is the wind speed; is the wire diameter; is the wire tension; is the Reynolds number, which is used to describe the fluid flow state. , , To improve the dynamic viscosity of ice and the accuracy of wind field simulation, is the air dynamic viscosity; is the wind speed frequency; is the natural frequency of the conductor; is the damping ratio, which reflects the degree of energy loss when the system vibrates; is the wind direction angle; is the dancing amplitude; Dance amplitude for safety; is the critical dancing amplitude; is the wind speed frequency; is the resonant frequency of the conductor; is the maximum frequency; is the minimum frequency; For the duration of the dance; To allow for the maximum dancing time; Standardized scoring of wire galloping risk.

[0034] Based on fluid mechanics and vibration theory, the coupling relationship between wind speed, wind direction, and the natural frequency of conductors is analyzed to establish an instability coefficient model for conductor galloping. Wind speed and direction are monitored using an ultrasonic anemometer, and conductor vibration frequency is acquired using an accelerometer. When wind speeds exceed 15 m / s or the vibration frequency approaches the conductor's resonant frequency, the galloping amplitude and duration are calculated to assess risk. This provides early warning of conductor galloping risks and reduces the probability of fatigue damage to the conductors due to vibration by adjusting transmission capacity or activating damping devices.

[0035] Using data collected by high-definition cameras as input, the model is trained using the random forest algorithm to output the risk probability of line failure caused by bird activities.

[0036] ; in: scoring bird activity risk; is the vertical distance between the bird's nest and the conductor; For safe distance; is the bird activity density; is the minimum activity density; is the maximum activity density; Standardized scoring of bird activity risk.

[0037] Based on image recognition and statistical analysis, a bird damage risk assessment model was established, setting safe distance thresholds between bird nests and wires and bird activity density thresholds. High-definition cameras and the YOLOv5 algorithm were used to identify bird populations and nesting behaviors in real time, focusing on wetlands and bird migration routes. Alerts were triggered when nests were detected within two meters of wires or when the number of birds exceeded a threshold. This enabled timely identification of potential bird damage hazards, and the use of bird repellent devices or manual intervention to reduce the risk of short-circuit failures, increasing the prevention success rate by over 70%.

[0038] The dust and pollution flashover risk model formula is as follows: ; ; ; ; in: The dust pollution index is an indicator that measures the degree of dust pollution on line equipment; is the concentration of fine particulate matter; is the standard concentration; is the humidity influence coefficient, which reflects the influence of humidity on the insulator flashover voltage; is a coefficient, an empirical parameter determined by experiment or data analysis; is the relative humidity; is the standard relative humidity; The pollution flashover voltage is the critical voltage at which the insulator flashes under pollution and humidity conditions; is the reference flashover voltage; 、 、 are empirical coefficients, which are parameters determined based on actual operating data in historical data and experimental fitting; is the surface dirtiness; is the minimum dust pollution index; is the maximum dust pollution index; is the optimal relative humidity; is the maximum relative humidity; is the minimum relative humidity; is the minimum pollution flashover voltage; is the maximum pollution flashover voltage; Standardized scoring of dust and pollution flashover risk.

[0039] Based on the flashover mechanism of contaminated insulators, analysis A flashover voltage prediction model was established based on the synergistic effects of PM2.5 concentration, air humidity, and insulator surface contamination. Laser dust sensors were used to monitor PM2.5 concentrations, while hygrometers were used to collect air humidity data. The salt density of the insulator surface was calculated to assess the impact of dust pollution and high humidity on flashover voltage. This model predicts the insulator flashover risk level, guiding operations and maintenance personnel to rationally schedule insulator cleaning cycles and reducing the incidence of pollution flashover accidents by over 50%.

[0040] The geological disaster risk model formula is as follows: ; in: is the surface displacement rate (mm / h), which is monitored in real time by the GNSS module; is the safety displacement threshold (usually 0.1-0.5mm / h); is the critical displacement rate (the threshold for triggering an early warning, such as 5 mm / h); is the peak acceleration (g value), collected by the acceleration sensor; The acceleration threshold that the device can tolerate (e.g., 0.3g); is a standardized score for geological hazard risk. When the value is ≥0.6 and lasts for more than 10 minutes, it is judged as a high-risk landslide.

[0041] Based on the coupling principle of surface displacement and acceleration, a dual-parameter risk assessment model, combining surface displacement rate and peak acceleration, was established to analyze the impact of geological hazards such as landslides and collapses on tower stability. The surface displacement rate is monitored in real time using a GNSS module, and the peak acceleration of the tower foundation is collected using an accelerometer. Both are compared with safety thresholds to calculate a risk score. (When the surface displacement rate exceeds the safety threshold and the peak acceleration exceeds the equipment tolerance, the risk level increases significantly.) This system provides real-time warnings for geological hazard risks. When the risk score exceeds 0.6 and persists for more than 10 minutes, it is identified as a high-risk landslide, triggering tower foundation reinforcement or emergency line inspections.

[0042] The equipment aging risk model formula is as follows: ; in: is the crack depth detected by ultrasonic testing (mm); The maximum crack depth allowed for the insulator (e.g. 3mm); is the crack growth rate coefficient (calibrated by accelerated aging test); is the operating time (years); is the metal thickness loss measured by the corrosion sensor ( ); is the critical corrosion thickness (e.g. 20% of the thickness of the metal fitting); is a standardized score for equipment aging risk. ≥0.7 or ≥50 When the device is determined to need immediate replacement.

[0043] Based on the theory of material fatigue and corrosion, a dual-index evaluation model for insulator crack growth and hardware corrosion was established to analyze the relationship between equipment aging and operating time. The depth of internal cracks in the insulator was detected using an ultrasonic flaw detector, and the metal thickness loss of the hardware was monitored using a corrosion sensor. Combined with the equipment's operating years, a comprehensive risk score for crack growth rate and corrosion degree was calculated (a high-risk state is determined when the crack depth approaches the maximum allowable value or the metal thickness loss exceeds the critical value). The aging risk level of the equipment is quantified. When the risk score is ≥0.7 or the metal thickness loss is ≥50 When the device is replaced, it will trigger the equipment replacement process to avoid line failures caused by broken insulators or broken hardware.

[0044] Risks are inherently physical phenomena (e.g., ice accumulation is condensation, lightning strikes are electrical discharges). Multi-physics coupled predictive models, based on known laws (e.g., conservation of energy, heat conduction), can explain the causes of risks, avoiding the "black box" limitations of purely data-driven models. For example, purely data-driven models may mistakenly associate high temperatures with wildfires, while physical models can distinguish "high temperature ≠ fire" (e.g., high temperatures without vegetation will not cause wildfires). Using model formulas, operations and maintenance personnel can understand why risks occur (e.g., "conductor stress exceeds specified limits due to heavy ice accumulation"), rather than relying solely on algorithmic results. Even in data-scarce scenarios (e.g., new lines), physical models can still predict risks through theoretical deduction, thus compensating for data deficiencies.

[0045] The specific process of the Quantum-Graph Attention Network (Q-GAT) fusion model is as follows: Mapping physical field characteristics into quantum states , enhanced features are obtained by measurement , combined with the graph attention mechanism , extract graph features and output node weights : ; ; in: For the The importance weight of the risk feature class satisfies .

[0046] The specific process of transferring the reinforcement learning fault diagnosis model is as follows: According to the policy gradient algorithm , output fault type decision probability matrix and the fault severity score vector The calculation process of the fault type decision probability matrix is ​​as follows: ; in: In state The fault type is judged as probability; The number of states covers various operating conditions during line operation, such as normal operation, increased icing, abnormal wind speed, etc. The number of fault types includes icing faults, wildfire faults, lightning faults and other typical power line faults.

[0047] The fault severity score vector calculation process is as follows: ; in: Score the severity of the fault corresponding to each risk, such as (Icing Fault Severity Rating), (Wildfire Fault Severity Score), etc.; The pre-set The fault severity benchmark value is determined based on historical fault data and expert experience, and its value range is [0,1]. A larger value indicates a higher fault severity.

[0048] When the system detects new area or new working condition data, the policy network parameters of the source domain (area or working condition with rich data and training experience) are adjusted by domain adaptation technology. Parameters of the target domain (new region or new operating condition) To perform fusion, the formula is ,in is the fusion coefficient, with a value range of [0,1], which is determined through experimental optimization to accelerate the learning speed of the target domain and improve the accuracy of fault diagnosis.

[0049] Traditional physical models struggle to handle complex nonlinear relationships (such as the coupling of multiple risks and the gradual aging of equipment). For example, the combined effects of ice cover, dust, and bird aggregation cannot be calculated using simple formulas. Instead, algorithms must be used to discover patterns in historical data. Quantum-graph attention networks can map multi-source data into quantum states (e.g., ice thickness and wind speed as qubits), extract features using quantum circuits, and then use graph attention mechanisms to analyze inter-tower relationships (e.g., whether risks from adjacent towers affect each other). This is similar to using quantum computers to analyze power grid topology, capturing hidden correlations in high-dimensional data (e.g., the relationship between ice cover and terrain elevation). Transfer reinforcement learning algorithms, on the other hand, learn optimal strategies through trial and error. For example, in an icing scenario, they try different ice-melting currents and optimize the strategy based on rewards (e.g., a decrease in ice cover rate). When encountering new scenarios, they transfer existing experience (e.g., migrating from area A to area B) to accelerate learning. This is like "an algorithm becoming an expert in risk management through 'experience accumulation'." Combining these two approaches can effectively reduce prediction errors and, when faced with new risks (e.g., novel bird nesting behaviors), quickly adjust recognition strategies without the need for manual reprogramming.

[0050] Based on the above content, a comprehensive risk value calculation formula is constructed to calculate the comprehensive risk value. Risk levels are classified according to the calculated comprehensive risk value. Relevant information is sent to relevant personnel through different channels according to the risk level and relevant operations are performed. The comprehensive risk value calculation formula is as follows: ; in: is a comprehensive risk value, reflecting the overall risk level of the power line, with a value range of [0,1]. A larger value indicates a higher risk. For the The combined weight of the risk class is determined by the analytic hierarchy process and entropy weight method; For the The standardized score of the risk class is calculated by the corresponding risk assessment formula and the value range is [0,1]; is the risk coupling coefficient, with a value range of [0.2, 0.5]. It is used to measure the degree of mutual influence between different risks and can be determined based on the correlation analysis of historical data; For risk and The correlation coefficient is obtained by calculating the Pearson correlation coefficient of the two types of risk scores, which reflects the degree of association between risks; The output of the Q-GAT model is The weight of the risk feature reflects the importance of the risk feature in the graph structure; is a regulating factor with a value range of [0.1, 0.3], which is used to adjust the impact of the transfer reinforcement learning fault severity score on the comprehensive risk value; The output of the transfer reinforcement learning model The fault severity score corresponding to the risk class.

[0051] The comprehensive risk value obtained based on the above calculation , divide the risk into five levels. Set multiple thresholds, When the risk level is low, use express; When the risk level is medium, use express; When the risk level is high, use express; When the risk level is urgent, use express; When the risk level is catastrophic, use The specific value of the threshold needs to be defined according to the specific environment. Different environments have different thresholds, and the single risk threshold also needs to be adjusted according to the actual situation. If it is not within this range, it means that the calculation is wrong and needs to be recalculated.

[0052] For example, the threshold settings in the coastal / industrial pollution area are as follows: When the risk level is When the risk level reaches a low level, the monitoring data needs to be archived for subsequent model training and optimization; inspections should be carried out in accordance with normal requirements; and system performance evaluation and maintenance should be carried out regularly to ensure stable operation of the system. When the risk level is Medium-risk level, the system will automatically generate routine maintenance work orders, arrange operation and maintenance personnel to conduct inspections and maintenance according to the plan; optimize inspection routes, use intelligent algorithms to plan the best inspection routes, improve operation and maintenance efficiency; strengthen online monitoring, and increase the frequency of data collection. When the risk level is High-risk level, the system pushes AR maintenance guides to operation and maintenance personnel, guides on-site maintenance through augmented reality technology; starts drone inspections to increase inspection frequency; closely monitors the line operation status, adjusts operating parameters in real time, and conducts local power outages when necessary. When the risk level is Emergency level: At this time, the system will give priority to using 5G slicing network to transmit monitoring data to ensure real-time data; dynamically adjust line de-icing, lightning protection and other protection strategies; reduce transmission capacity to avoid fault expansion; arrange professional operation and maintenance personnel to rush to the scene with portable detection equipment to conduct emergency investigation and preliminary treatment. When the risk level is Disaster level: At this time, the system automatically triggers trip protection, cuts off the power supply of the faulty line; starts satellite relay communication to ensure the transmission of critical data; dispatches drone fire-fighting clusters, repair robots and other equipment; and notifies the emergency command center to organize cross-regional professional repair teams to carry out large-scale rescue and disaster relief work.

[0053] Comprehensive assessments can effectively reduce the probability of misjudgments, and tiered early warnings can more effectively allocate resources (such as increased drone inspections). They can also reduce costs during low-risk periods and avoid overprotection. Tiered early warnings can also shorten risk response time and improve grid reliability. This solution forms a closed "monitoring-analysis-response" loop through a process that includes data collection (comprehensive perception) → spatiotemporal calibration (data alignment) → physical modeling (analysis of patterns) → algorithm optimization (intelligent decision-making) → tiered response (precise handling), ensuring safe grid operation.

[0054] Multi-source heterogeneous data fusion and spatiotemporal calibration technology: A monitoring network is constructed using multiple sensors, including laser ranging, satellite remote sensing, and fiber-optic temperature measurement. Combining the BeiDou spatiotemporal benchmark (B1 / B3 frequencies) with the Kalman filter algorithm, this achieves microsecond-level time calibration and meter-level spatial calibration. An improved DTW algorithm introduces spatiotemporal constraints to achieve microsecond-level time alignment of multi-source data. These multiple sensors cover eight risk categories (icing, wildfires, lightning, etc.), providing a comprehensive inspection of power lines and preventing any missed detections by a single sensor. For example, laser ranging to monitor ice thickness (accuracy ±1mm) combined with distributed fiber-optic temperature measurement (temperature gradient ±0.5°C / 100m) allows for simultaneous analysis of the impact of ice on conductor temperature, rather than monitoring a single parameter in isolation. BeiDou calibration addresses spatiotemporal deviations in raw sensor data (e.g., ionospheric delays up to 500ns and positioning errors of ±500 meters), ensuring that ice data from tower A is not mistakenly associated with tower B, thus preventing false alarms caused by misidentification. For example, in lightning location, the calibration results in a lightning strike location error of less than 500 meters, enabling precise location of faults and shortening repair times. Calibrated data can accurately trace the spatiotemporal trajectory of risk occurrence (e.g., a sudden stress change on a tower at a specific moment), providing reliable input for physical models (such as conductor thermal-mechanical coupling models), reducing prediction errors to less than 3%.

[0055] Collaboration between the Multiphysics Coupling Model and the Quantum-Graph Attention Network: The multiphysics coupling model builds mathematical models based on energy conservation and heat conduction laws. For example, in the conductor thermomechanical coupling model, the conductor is divided into 100 units using the finite element method to collaboratively predict temperature, stress, and ice thickness. The Q-GAT algorithm maps multi-source data into quantum states, extracts features in parallel using quantum circuits, and combines this with a graph attention mechanism to analyze tower topological relationships (e.g., risk coupling between adjacent towers). The algorithm then outputs risk feature weights, enabling the discovery of hidden associations in high-dimensional data (e.g., the nonlinear relationship between ice cover and altitude and wind speed). The physical model provides theoretical support for the causes of risks (e.g., "excessive ice cover causes excessive conductor stress"), avoiding the "black box" limitations of purely data-based algorithms. Q-GAT optimizes model parameters by learning from historical data (e.g., over 100,000 ice cover records), improving prediction accuracy in complex scenarios (e.g., multiple coupled risks). For example, in wildfire models, physical models distinguish "high temperature without vegetation ≠ fire." Q-GAT, by analyzing satellite thermal imaging data and vegetation distribution, improves fire point identification accuracy to 95%. Traditional models struggle with high-dimensional nonlinear problems (such as the combined effects of dust concentration, humidity, and pollution on flashover voltage). Through quantum feature extraction and graph analysis, Q-GAT can identify the exponential relationship between dust pollution index and flashover voltage, reducing flashover risk prediction errors by 40%. For new lines or data-scarce scenarios, physical models provide basic predictions through theoretical derivation. Q-GAT accelerates model convergence through transfer learning (e.g., reusing data from other regions), reducing reliance on local data and improving generalization capabilities.

[0056] An adaptive decision-making system driven by transfer reinforcement learning: Using a policy gradient algorithm, the system uses fault characteristics (such as icing rate and stress value) as states and optimizes the decision-making policy using a reward function (e.g., a reward of +10 for a decrease in icing rate). The system outputs a fault type probability matrix and severity score. A domain-adaptive transfer method is also employed to rapidly adapt to new environments by fusing policy network parameters from the source domain (e.g., the North China Plain) and the target domain (e.g., the Sichuan-Tibet Plateau). For example, icing response strategies in the plateau region can be adapted by transferring de-icing experience from the plains, improving learning efficiency by 60%. Traditional systems rely on fixed threshold warnings (e.g., wind speeds ≥ 15 m / s trigger a sway warning), while DRL dynamically adjusts thresholds through continuous interactive learning. For example, during coastal typhoon season, DRL can automatically lower the sway warning threshold to 12 m / s based on historical data, triggering a response two hours in advance and reducing the accident rate. In comprehensive risk assessments, DRL balances safety and cost-effectiveness. For example, when icing risk and equipment aging risk coexist, DRL can automatically select a "melt-first + simultaneous maintenance" strategy to avoid wasting resources associated with single-risk response. Faced with new types of bird nesting behavior (such as bird nests that have never been recorded and are less than 2 meters away from the wire), DRL updates the recognition model through small sample learning (only 50 new data points are required), without the need for manual reprogramming, and the response speed is increased by 90%.

[0057] For example, in winter, on a certain tower on a certain transmission line in Sichuan and Tibet, the laser ranging sensor measured the ice thickness at 12:00:00 on a certain day. , measured at 13:00:00 , conductor stress (Safety threshold ), conductor temperature (Optimum temperature ), On the same day, from 10:00:00 to 11:00:00 in the morning, satellite remote sensing detected the surface temperature , vegetation humidity (vegetation is moist in winter), wind speed (no wind), Fire Danger Index (below the warning threshold of 50), score On that day, the lightning location system detected 0 lightning strikes. The average number of lightning strikes per year is 12, with winter accounting for <5%. At 10:00:00-11:00:00 on the same day, the ultrasonic anemometer detected the wind speed (< Safety threshold), score On that day, only a single bird was detected flying by, no bird nests were found, and the bird activity density was Birds / hectare (below the threshold of 10 birds / hectare), score The laser dust sensor detected Concentration 50 (<35 Excellent threshold), relative humidity ,score On that day, the GNSS module detected the surface displacement rate (<0.5 Safety threshold), the acceleration sensor detects a peak (less than 0.3 threshold), score On that day, the ultrasonic flaw detector detected the depth of the insulator crack. (<3 Allowable value), the corrosion sensor detects the metal thickness loss (< Safety value), score . Risk Weight , , the remaining risk is approximately equal to 0, the coupling coefficient , comprehensive score ( Low risk). Required measures include: arranging nighttime infrared temperature checks after ice has melted, including equipment aging risks in the next month's maintenance plan, and prioritizing the replacement of hardware nearing critical values.

[0058] The embodiments disclosed in the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. The embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the above-mentioned functions defined in the method of the present application are executed. It should be noted that the computer-readable medium mentioned above in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of an electrical, magnetic, optical, electromagnetic, infrared segment or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more wire segments, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including, but not limited to, an electromagnetic signal, an optical signal, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, electrical wire, optical fiber cable, RF, etc., or any suitable combination thereof.

[0059] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or portion of code that contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or operations, or can be implemented using a combination of dedicated hardware and computer instructions.

[0060] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the implementation methods of the present invention may be subject to any deformation or modification.

Claims

1. A power line alarm method with AI function, characterized in that: The method comprises: Use multiple sensors to collect multi-source data, receive multi-source data information through the Beidou dual-mode module, and use the Kalman filter algorithm to perform spatiotemporal calibration on the multi-source data information; Based on the collected multi-source data information, a multi-physics field coupling prediction model is constructed, and standardized scores for different risk types are calculated through the multi-physics field coupling prediction model. The multi-physics field coupling prediction model includes a conductor thermal-mechanical coupling model, a wildfire spread dynamic model, a lightning risk prediction model, a conductor dancing prediction model, a bird activity risk model, a dust and pollution flashover risk model, a geological disaster risk model, and an equipment aging risk model. The quantum-graph attention network fusion model is used to calculate the weights of different risk features, and the fault severity scores corresponding to different types of risks are calculated through the transfer reinforcement learning fault diagnosis model; A comprehensive risk value calculation formula is constructed by combining the standardized scores of different risk types, the weights of different risk characteristics, and the fault severity scores corresponding to different risk types. The comprehensive risk value is calculated using the comprehensive risk value calculation formula. Risks are classified into different levels according to the threshold range of the comprehensive risk value, and different warning information and processing strategies are pushed to different responsible personnel according to different risk levels.

2. The power line alarm method with AI function according to claim 1, characterized in that: The spatiotemporal calibration comprises the following steps: The B1 and B3 frequency signals are received through the Beidou dual-mode module. The Kalman filter algorithm is used to deduct the error between the sensor's original time and the Beidou standard time, the ionospheric delay correction value, and the tropospheric delay correction value to achieve time calibration. The positioning error vector of the sensor's original spatial coordinates is corrected to achieve spatial calibration. The improved DTW algorithm is used to introduce time and space constraint factors to achieve data alignment.

3. The power line alarm method with AI function according to claim 2, characterized in that: The conductor thermal-mechanical coupling model constructs a model formula by collaboratively predicting changes in temperature, stress, and ice thickness, divides the unit conductor into 100 units, and uses the finite element method to solve it; the wildfire spread dynamic model uses the principle of heat conduction and fire risk assessment indicators to analyze the impact of surface temperature, vegetation humidity, and wind speed on the fire spread rate, and uses the Kalman filter algorithm to update the model parameters; the lightning risk prediction model establishes an equipment damage probability model through a Poisson process combined with the equipment tolerance current threshold to predict the probability of equipment damage caused by lightning strikes; the conductor galloping prediction model is based on fluid mechanics and vibration theory, analyzes the coupling relationship between wind speed, wind direction and the conductor's natural frequency, and establishes an instability coefficient model for conductor galloping.

4. The power line alarm method with AI function according to claim 3, characterized in that: The bird activity risk model is based on image recognition and statistical analysis, sets the safe distance threshold between bird nests and conductors and the bird activity density threshold, and establishes a bird damage risk assessment model; the dust and pollution flashover risk model is based on the flashover mechanism of polluted insulators, combined with The geological disaster risk model is based on the principle of coupling between surface displacement and acceleration, and a dual-parameter risk assessment model of surface displacement rate and acceleration peak is established; the equipment aging risk model is based on the theory of material fatigue and corrosion, and a dual-index assessment model of insulator crack extension and hardware corrosion is established.

5. The power line alarm method with AI function according to claim 4, characterized in that: The process of obtaining weights of different types of risk characteristics is as follows: The physical field characteristics of multi-source data are mapped into quantum states, and superposition states and entangled states are generated through quantum gate operations to form a quantum feature space; the quantum states are measured, nonlinear features are extracted, and enhanced feature vectors are generated; the power network is abstracted into a graph structure, where nodes represent poles or equipment, and edges represent physical connections or risk propagation relationships between nodes. The association weights between nodes are calculated through the attention mechanism; the neighbor node features are aggregated based on the attention weights, the current node representation is updated, and the weights of different types of risk features are output.

6. The power line alarm method with AI function according to claim 5, characterized in that: The process of transferring the reinforcement learning fault diagnosis model is as follows: The operating conditions of power lines are converted into a computable state space, and a set of fault handling actions is preset for fault handling. A deep neural network is used to construct a policy function, which inputs the state, outputs the action probability distribution, sets reward rules, updates the policy network parameters based on the policy gradient algorithm, and outputs the fault type decision probability matrix and the fault severity score vector. When the system accesses new regional data, the policy network parameters of the source domain and the target domain are fused through domain adaptation technology.

7. The power line alarm method with AI function according to claim 6, characterized in that: The specific formula for calculating the comprehensive risk value is as follows: ； in: is the comprehensive risk value; For the The combined weight of the risk class; For the Standardized scoring of class risks; is the risk coupling coefficient; For risk and The correlation coefficient of The output of the Q-GAT model is Class risk characteristic weights; is the regulating factor; The output of the transfer reinforcement learning model Fault severity score corresponding to the risk class; For risk category.

8. The power line alarm method with AI function according to claim 7, characterized in that: The specific process of the grading is as follows: when When the risk level is low , archive monitoring data, carry out inspections according to normal requirements, and regularly evaluate and maintain the system's performance; When the risk level is medium , the system will automatically generate routine maintenance work orders, arrange operation and maintenance personnel to conduct inspections and maintenance as planned, strengthen online monitoring, and increase the frequency of data collection; When the risk level is high The system pushes AR maintenance guides to operation and maintenance personnel, guides on-site maintenance through augmented reality technology, activates drone inspections, increases inspection frequency, closely monitors line operation status, adjusts operating parameters in real time, and conducts local power outage inspections when necessary; When the risk level is urgent The system prioritizes the use of 5G slicing networks to transmit monitoring data, dynamically adjusts line protection strategies, reduces transmission capacity, and arranges professional operation and maintenance personnel to rush to the site with portable testing equipment to conduct emergency investigations and preliminary processing; When the risk level is catastrophic The system automatically triggers the tripping protection, cuts off the power supply of the faulty line, starts satellite relay communication, dispatches drone fire-fighting clusters and repair robots, and notifies the emergency command center to organize a cross-regional professional repair team to carry out large-scale rescue and disaster relief work.

9. A power line alarm system with AI function, characterized in that: The system is used to execute the power line alarm method with AI function as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the power line alarm method with AI function as described in any one of claims 1 to 8.

Images